This research explores the potential of Machine Learning (ML) to enhance wireless communication networks, specifically in the context of Wireless Smart Grid Networks (WSGNs). We integrated ML into the well-established Routing Protocol for Low-Power and Lossy Networks (RPL), resulting in an advanced version called ML-RPL. This novel protocol utilizes CatBoost, a Gradient Boosted Decision Trees (GBDT) algorithm, to optimize routing decisions. The ML model, trained on a dataset of routing metrics, predicts the probability of successfully reaching a destination node. Each node in the network uses the model to choose the route with the highest probability of effectively delivering packets. Our performance evaluation, carried out in a realistic scenario and under various traffic loads, reveals that ML-RPL significantly improves the packet delivery ratio and minimizes end-to-end delay, making it a promising solution for more efficient and responsive WSGNs.

Vehicular networks support intelligent transportation system (ITS) to improve drivers' safety and traffic efficiency on the road by exchanging traffic-related information between vehicles and also between vehicles and infrastructure. Routing protocols that are designed for vehicular networks should be flexible and able to adapt to the inherent dynamic network characteristics of these kind of networks. Therefore, there is a need to have effective vehicular communications, not only to make mobility more efficient but also to reduce collateral issues such as pollution and health problems. Nowadays, the use of machine learning (ML) algorithms in wireless networks are on the rise, including vehicle networks that can benefit from possible data-driven predictions. This work aims to contribute to the design of a smart ML-based routing protocol for vehicular ad hoc networks (VANETs) used to report traffic-related messages in urban environments. We propose a new ML-based forwarding algorithm to be used by the current vehicle holding a given packet to predict which vehicle within its transmission range is the best next-hop to forward that packet towards its destination. Our algorithm is based on a neural network designed from a dataset that contains data records that are captured during simulated urban scenarios. Simulation results show how our ML-based proposal improves the performance of our multimetric routing protocol for VANETs in urban scenarios in terms of packet delivery probability. The performance evaluation of MPANN shows packet losses lower than 20% (and average packet delays below 0.04 ms) for different vehicles' densities, in completely new scenarios but of similar complexity than the Barcelona scenario used to train the model. Even for much more complex scenarios (with narrow curvy streets), our proposal is able to reduce the packet losses in 20% with respect to the multimetric routing protocol as well as the average packet delays in 0.04 ms.

Smart Grid (SG) networks include an associated data network for the transmission and reception of control data related to the electric power supply service. A subset of this data network is the SG Neighborhood Area Network (SG NAN), whose objective is to interconnect the subscribers' homes with the supplier control center. The data flows transmitted through these SG NANs belong to different applications, giving rise to the need for different quality of service requirements. Additionally, other subscriber appliances could use this network to communicate over the Internet. To avoid network congestion, as well as to differentiate the quality of service (QoS) received by the different data flows, a congestion control mechanism with traffic differentiation capabilities is required. The main contribution of this work is the proposal of a new congestion control mechanism based on machine learning techniques to try to guarantee the different QoS requirements to the different data flows. A main problem when applying machine learning techniques is the need for datasets to be used in the training steps. In this sense, a second contribution of this article is the proposal of a method to generate such datasets by means of simulation techniques. The proposed mechanism is then evaluated in the context of a wireless SG NAN. The nodes of this network are the subscriber's smart meters, which in turn perform the function of concentrating the data traffic sent and received by the rest of the home appliances. Besides, different machine learning classification methods are taken into account. The evaluation carried out shows significant improvements in terms of network throughput, transit time, and quality of service differentiation. Finally, the computational cost of the algorithms used in this proposal has also been evaluated, using real low-cost IoT hardware platforms.

Modern electrical systems are evolving with data communication networks, ushering in upgraded electrical infrastructures and enabling bidirectional communication between utility grids and consumers. The selection of communication technologies is crucial, where wireless communications have emerged as one of the main benefactor technologies due to its cost-effectiveness, scalability, and ease of deployment. Ensuring the streamlined transmission of diverse applications between residential users and utility control centers is crucial for the effective data delivery in smart grids. This paper proposes a congestion control mechanism tailored to smart grid applications using unreliable transport protocols such as UDP, which, unlike TCP, lacks inherent congestion control, presenting a significant challenge to the performance of the network. In this article, we have exploited a reinforcement learning (RL) algorithm and a deep Q -neural network (DQN), to manage congestion control under a UDP environment. In this particular instance, the DQN model learns on its own from interactions with the environment without the need to generate a dataset, rendering it apt for intricate and dynamic scenarios within smart grid communications. Our evaluation covers two scenarios: (i) a grid-like configuration, and (ii) urban scenarios considering the deployment of smart meters in the cities of Montreal, Berlin and Beijing. These evaluations provide a comprehensive examination of the proposed DQN-based congestion control approach under different conditions, showing its effectiveness and adaptability. Conducting a comprehensive performance assessment in both scenarios leads to improvements in metrics such as packet delivery ratio, network throughput, fairness between different traffic sources, packet network transit time, and QoS provision.